Bayesian ensemble methods for predicting ground deformation due to tunnelling with sparse monitoring data

Numerous analytical models have been developed to predict ground deformations induced by tunneling, which is a critical issue in tunnel engineering. However, the accuracy of these predictions is often limited by errors and uncertainties resulting from model selection and parameter fittings, given the paucity of monitoring data in field settings. This paper proposes a novel approach to estimate tunnelling-induced ground deformations by applying Bayesian model averaging to several representative prediction models. By accounting for both model and parameter uncertainties, this approach enables more realistic predictions of ground deformations than individual models. Specifically, our results indicate that the Gonzalez-Sagaseta model outperforms other models in predicting ground surface settlements, while the Loganathan-Poulos model is most suitable for predicting subsurface vertical and horizontal deformations. Importantly, our analysis reveals that when monitoring data are sparse, model uncertainties may contribute up to 78.7% of the total uncertainties. Thus, obtaining sufficient data for parameter fitting is crucial for accurate predictions. The proposed method in this study offers a more realistic and efficient prediction of tunnelling-induced ground deformations.